12/05/14 CAS2014 ADCs & DACs 1

CERN Accelerator SchoolADCs and DACs

(Analogue to Digital and Digital to Analogue Converters)

John Pickering, Metron Designs Ltd, Norwich, UK

With grateful acknowledgment to John Pett (retired from CERN)

12/05/14 CAS2014 ADCs & DACs 2

ADCs & DACs General Use

• Interface between Analogue & Digital environments• Trend is “Digital to the Antenna”• Most ADCs contain DACs in a feedback loop• Delta-Sigma ADCs use both ADCs and DACs in the loop

12/05/14 CAS2014 ADCs & DACs 3

ADC DAC Subjects

• Terms, nomenclature, standards and specs.• Common ADC types

– Will mainly discuss DACs within ADC types– Flash– Pipeline– Successive Approximation Register (SAR)– Charge Balance, DVMs– Delta Sigma

• Applicability of types• Off the shelf “state of the art”• Some application problems & solutions• The future for accelerators?

12/05/14 CAS2014 ADCs & DACs 4

Terms, Nomenclature & Specs

• Many terms relate to AC measurement– Eg “SINAD”, “Over-sampling”– Standards IEEE1241 and DYNAD are primarily for Dynamic

performance• Methods & draft standards for the DYNamic testing of Analog to Digital

converters, European project SMT4-CT98-2214

• Many misunderstood terms and specs.– Eg. “resolution”, “linearity”, “noise”– Useful general standard is IEC 60748-4

• Now updated to include dynamic performance by incorporating IEEE1241 and DYNAD to give IEC 60748-4-3

12/05/14 CAS2014 ADCs & DACs 5

Terms, Nomenclature & Specs

• Resolution…the first quantity to know!– “Smallest repeatably discernable increment”, ie resolvable

increment.• IEC 60748-4 says, for ADCs, “nominal value of the step width” THIS IS

QUITE WRONG IN MY VIEW.• Very often applied as “N” bits or “digits” where it is the theoretical, error

free, noise free, digital scale increment• The term resolution is not even defined in IEEE1241 and DYNAD!!

• Noise….the second?– ADCs generally specify in RMS % of the bipolar scale length.

DVMs in peak % of unipolar scale length.• This is a 6:1 difference!!

12/05/14 CAS2014 ADCs & DACs 6

Terms, Nomenclature & Specs

• A perfect Bipolar ADC transfer function

Input Signal

Output Code

001

000

010

100

011

101

110

111

-111

-110

-100

-101

-011

-010

-001

0-1-2-3-4-5-6-7-8 1 2 3 4 5 6 7 8

LINEAR BIPOLAR ADC

Step width = q

“q” is the quantisation width

q =Full Scale Input Analog

Number of steps

Input Signal

Output Code

001

000

010

100

011

101

110

111

-111

-110

-100

-101

-011

-010

-001

0-1-2-3-4-5-6-7-8 1 2 3 4 5 6 7 8

NOISY ADC

Note blurred or widenedswitch thresholds

12/05/14 CAS2014 ADCs & DACs 7

Terms, Nomenclature & Specs

• Quantisation Error– “q” is the width of the LSB in

linear ADCs

– As input signal is increased the error has a saw-tooth distribution as it steps.

– Peak error in an ideal ADC is q/2.

– Standard deviation is q/12

– Dynamically, “quantisation noise” is therefore ……q/12 RMS

Input Signal

OutputCode

001

000

010

100

011

101

110

111

0 1 2 3 4 5 6 7 8

ADC Quantisation Error

+q/2-q/2

=q

12

12/05/14 CAS2014 ADCs & DACs 8

Terms, Nomenclature & Specs

• DNL…Differential Non-Linearity

– Defined as the difference between ideal and actual step width

– Significant problem in SAR ADCs and ladder DACs at major code transitions

– Can be many bits.

– Think of as a sudden step change in output.

Input Signal001

000

010

100

011

101

110

111

0 1 2 3 4 5 6 7 8

ADC with DNL ERROR

OutputCode

+1/2 q

-1/4 q

Defined as: “Difference betweenideal and actual step width”

q

1½ q

Monotonic, nomissing codes

12/05/14 CAS2014 ADCs & DACs 9

Terms, Nomenclature & Specs

• Missing Codes

– Extreme DNL, especially if > 1 LSB, can lead to missing codes and non-monotonic performance.

Input Signal001

000

010

100

011

101

110

111

0 1 2 3 4 5 6 7 8

ADC with DNL ERROR

Missing code

OutputCode

+3/4 q

-1/4 q

And missing code

12/05/14 CAS2014 ADCs & DACs 10

Terms, Nomenclature & Specs

• Non-Monotonic– Usually taken to mean that

different inputs can produce the same output code…ie two codes.

– In this unusual example all codes are unique although one is wrong.

– Definition is that the differential output takes a negative value for +ve differential input (and vice versa).

Input Signal

OutputCode

001

000

010

100

011

101

110

111

0 1 2 3 4 5 6 7 8

NON-MONOTONIC ADC

Defined as: “having a negative differential between output codes”

This case hasno missing codes

12/05/14 CAS2014 ADCs & DACs 11

Terms, Nomenclature & Specs

• INL…Integral Non-Linearity– Usually the limiting

performance for high resolution, low speed ADCs

– Is actually an accumulation of DNL errors over many steps.

– “Ideal transitions” are assumed to be on a straight line. The line may be between calibrated end-points which is the worst case but for specmanship reasons, a best fit regression line is sometimes used

Input Signal

OutputCode

001

000

010

100

011

101

110

111

0 1 2 3 4 5 6 7 8

ADC with INL errorDNL is < ½q but

INL>q

Defined as: “Extreme deviationfrom ideal transition”

12/05/14 CAS2014 ADCs & DACs 12

Terms, Nomenclature & Specs

• Dynamic terms mainly used for AC– SNR, Signal to Noise Ratio

• Ratio in dB of the RMS of a full excursion sine wave to the RMS of all of the noise including quantisation noise (with that signal applied and therefore specified).

– SFDR, Spurious Free Dynamic Range.• Ratio in dB of the value of a full scale sinewave signal to the highest spurious or

harmonic tone generated by the ADC’s operation

– SINAD, Signal to Noise and Distortion.• As SNR but including all distortion terms, treated as noise

– Nef or “ENOB”, effective number of bits.

• The effective resolution of the converter under the above conditions.Nef= (SINADdBFS-1.76dB)/6.02

– “Oversampling”…A confusing term for DC!!• Sampling faster than the Nyquist rate, ie faster than 2X the signal bandwidth. This is an

inadequate description for DC since, by definition it is infinite! In reality “oversampling” ADCs and DACs improve measurement of AC or DC by averaging multiple samples in the presence of dither or noise.

12/05/14 CAS2014 ADCs & DACs 13

Flash ADCs

• At its simplest a single comparator……….

– In practice one comparator per level or 2n-1 for “n” bits.

– A 16 bit converter therefore needs 65,535 comparators with thresholds set to an accuracy of .0015% of FS!

– However, at low resolution is often a sub-component in other types of ADC architecture.

12/05/14 CAS2014 ADCs & DACs 14

Flash ADCs, characteristics

• The flash ADC is one of the few types that does not include a DAC in a feedback loop– This means that there is the shortest possible path between

signal and output, ie between Analogue and Digital.• Flash converters are very fast, the fastest available.

• They have very short delays, (latency or group delay) as well as high bandwidth.

• They are therefore relatively easy to integrate into feedback loops.

• They are used within both Pipeline and - converters.

12/05/14 CAS2014 ADCs & DACs 15

Pipeline Converters

• Pipeline ADCs are multi-stage flash converters• If the resolution of each stage is the same, say n bits, then the result is nN

bits where N is the number of stages. In practice 1 bit or more is “lost” in each stage to allow calibration correction.

SAMPLE&

HOLD

FLASHADCn-bit

DAC

G~n

DIGITAL PATH

ANALOGUE PATH

+_

++

STAGE1

SAMPLE&

HOLD

FLASHADCn-bit

DAC

G~n

DIGITAL PATH

ANALOGUE PATH

+_

++

STAGEN

Principle of Pipeline ADC with n bitsN

12/05/14 CAS2014 ADCs & DACs 16

Pipeline Converters, characteristics

• Commercial Converters…

– Are typically 10 to 16 bits, 5MHz to 250MHz bandwidth

– Are clocked at 10MHz to 1GHz

– “Pipeline” 3 to 16 stages requiring one clock per stage.• This results in significant delay which can be high compared

with the bandwidth. This delay can be very problematic when they are used within feedback loops. eg a 10ns conversion rate can have a 160ns delay!!!

12/05/14 CAS2014 ADCs & DACs 17

Successive Approximation, SAR

• SAR ADCs use ladder DACs in a feedback loop to iterate towards a null balance between the signal and the DAC output.

• The ADC in the diagram below can be a simple comparator or a Flash Converter. The DAC may be resistive or capacitive.

12/05/14 CAS2014 ADCs & DACs 18

SAR, operation and characteristics

• Typically the converter divides the scale in two and tests successively for the DAC > or < the signal. Each test sets a “1” or “0” in the register at the appropriate weight

– Capable of higher resolution than pipeline at somewhat lower speed.

– Since it uses a binary ladder network, high order bit transitions can cause very bad DNL

Clock

Comparatorresult

MSB

LSB

DACoutput

FS

t

Result

1

0

0

1

Start Ends

10

10

10

10

10

Input Signal

0

12/05/14 CAS2014 ADCs & DACs 19

Charge Balance, Dual-slope

• Dual Slope A-D Sequence and Schematic

• Charge Balance equation Charge integrator with signal (Isig)

for fixed period Ts

Discharge for variable period tr with known reference (Iref)

 

t =n .tr r c

Sig

N ull

T =N .ts s c

To C ounter(clocked w ith t )c

R efrrefssig tITI

tcnItNI rrefcssig

s

refrsig T

ItI

s

refrsig N

InI

12/05/14 CAS2014 ADCs & DACs 20

Charge Balance, Multi-slope

• The Multi-slope makes two key improvements:– Removes “glugs” of reference charge allowing a smaller integrator cap and

thus higher gain to reduce effect of null detector noise. – Ends at a slow (but known) discharge rate whilst counting at reduced digital

significance thus improving resolution by “K” without a faster clock.

 I + Isig ref.5us

T r

T r te =n .te c

tr trt =n .tr r c

T =N .ts s c

M ore T cyclesr

Endof

Conversion

12/05/14 CAS2014 ADCs & DACs 21

Charge Balance, characteristics

• Most DVMs use Charge balance ADCs– But tend not to use ADC terminology and are specified in terms of near worst

case measurement errors. They also treat each polarity as having its own “FS”. They are optimised for DC so speed is slow.

• DVM specifications expressed as A-D “bits”– The following table gives ADC equivalent specs and is for a high end DVM,

however, this is an 18 year old design!

Specification

Description

DVM data-sheet specification

ADC “bit” specification

Nominal resolution: 8½ digits 28+ bits

Real (2) resolution 7½ digits 24+ bits

Integral Non-linearity (INL) 0.1ppm (1X10-7) 23 bits

Differential Non-linearity No spec. “perfect” 28 bits

12/05/14 CAS2014 ADCs & DACs 22

Charge Balance v Sigma-Delta

• Charge Balance “Rule”

– Charge IN = Charge OUT

– The best possible resolution is the smallest discrete charge that can be determined, compared with the total charge.

– This normally means the ratio of tc to Ts but in multi-slope this is extended by reducing the final charge current.

• The CERN Sigma-Delta

– A Sigma-Delta ADC was designed at CERN for controlling magnet currents to 1ppm. (ADDA 1999).

– It uses a 1 MHz “1-bit” topology with a digital filter that settles to 1ppm in 1000 clocks:- thus only 1000 quanta so:

Charge in Charge out!!

– It has been proven to reliably resolve 1ppm!

12/05/14 CAS2014 ADCs & DACs 23

Sigma-Delta

• Most Sigma-Delta ADCs use a “1-bit” DAC – The CERN architecture is shown below:

Note that DACs based on the same principle simply “swap” over the digital feedback and analogue input paths to give analogue feedback and a digital input bitstream that can be optimised for best response.

b u f f e r 3 i n t e g r a t o r s c o m p a r a t o r

L O G I C

D A C

“ 1 - b i t ” o / p

b i t s t r e a m

v i a f i b r e -

o p t i c l i n k

i n p u t

1 b i t

1 M H z

C L K

R E F .

D i g i t a l

F i l t e r

( F P G A )

o u t p u t

v a l u e

+_

12/05/14 CAS2014 ADCs & DACs 24

Sigma-Delta, 1-bit, 200 clocks

• Excel Simulation Output for 1 bit , 4X50 dig filter– The plot below shows the output of the last three filter registers over the last

140 out of 210 clocks. The Y scale is +/- 0.5%FS.

0.76

0.762

0.764

0.766

0.768

0.77

clock# (last 130 of 200)

Va

lue

fo

r e

ac

h s

um

ou

t re

gis

ter

(de

tail

)

sum2

sum3

sum4: Output

input value

1 bit Delta-Sigma 3rd order 4X50 filter settling(Equivalent Charge Balance Resolution 0.5%)

Note fine resolution of sum4 relative to 1 part in 200 !!

Fig 3

In this time dual-slope would only achieve a resolution of 1/200, ie 0.5%!

12/05/14 CAS2014 ADCs & DACs 25

Sigma-Delta characteristics

• Commercial Integrated Circuit (IC) converters– Mainly designed for audio frequency AC

• Therefore specs tend not to be DC & LF “friendly”• High manufacturing volume means low cost

– Bandwidths are DC to 1MHz• Noise-Resolution trade-off is readily available• Considerable delay “latency” in dig filter• But the digital filter can be tapped at different points and sampled

concurrently. This may allow “feed-forward” techniques to be used in loop controlled systems

– DC performance is usually limited…look carefully!• Although “24 bit” devices are claimed, the stability and linearity is really

in the 18 bit region for the best “merchant semiconductor” IC devices. INL is the limitation.

12/05/14 CAS2014 ADCs & DACs 26

High Res sampling v DS

• Hi Res sampling, eg Successive Approximation requires analog anti-aliasing filtering which, when viewed as part of the adc “black box” makes it indistinguishable from DS with digital filtering

In

Hi_res sampling

In

DS sampling

Hi_resADC

DSADC

DigFilt

Deci-mator

These two examples can be entirely equivalent…except that the digital filter can be much more complex and is inherently matched between channels. The DS converter still has an anti-aliasing filter but this is at the over-sampling rate so can be much simpler. So the latency of the DS actually represents analog latency in the slower anti-aliasing filter of the Hi-res ADC.

12/05/14 CAS2014 ADCs & DACs 27

Sigma-Delta characteristics

• Problem Characteristics– “Idle tones”

• Bit patterns can repeat at low frequency if the loop does not “randomise” enough, a particular tendency in 3rd order or below.

• These are called tones because they show up as low level 1Hz-300 Hz lines in an FFT analysis of output.

• They are, at least in part, due to unwanted feedback in the modulator. PCB layout and supply filtering is critically important.

– “Sticky Zero”• Perhaps can be considered as a DC “idle tone” this shows up as a

tendency to maintain a symmetrical, zero condition, bit pattern in the modulator when operating just off zero. Synchronous noise, particularly in reference switching edges, is often the culprit.

12/05/14 CAS2014 ADCs & DACs 28

Choosing the right type

• The following table attempts to “score” the suitability of a technology against application requirements

\ADC type

Requirement\Flash Pipeline SAR Charge

Balance

Sigma-

Delta

Throughput: excellent v.good good poor fair

Bandwidth: excellent excellent v.good v.poor fair

Resolution: poor good v.good excellent excellent

Latency/Hz: excellent fair v.good poor fair

Multiplexing: excellent poor v.good fair poor

Linearity/bit: v.good good fair v.good v.good

Comments: Power!

Cost

v.fast

clock

DNL

stability

DC

only

Easy

Anti-alias

12/05/14 CAS2014 ADCs & DACs 29

Choosing the right spec.

• Choice of ADC– Internal/External Reference

• After choosing the type of ADC there are important specification subtleties to consider

• Most IC ADCs and DACs use low voltage CMOS technology and therefore internal references, if offered, are based on bandgap technology. These have high 1/f noise so an external Zener based reference may give significantly better performance.

– Bipolar Operation• Many ADCs/DACs are designed to be unipolar and there can be some difficulty or

performance degradation in converting to bipolar.– Overloads

• Look at how cleanly overloads are dealt with and the consequent system implications.

– Linearity• INL is often ORDERS worse than the “bit” spec quoted

– Latency• Pipeline converters can have 16 sample delays!

12/05/14 CAS2014 ADCs & DACs 30

Commercial State of the Art

– Not all the parameters at the same time!!

Semiconductor ADC Technology

Claimed Resolution

Real Resolution

Highest Bandwidth

Latency /Settling

INL, bits

           

Flash 8 8 1GHz 2ns 8

Pipeline 16 15 250MHz 100ns 14

SAR 20 20 10MHz 100ns 19

Delta-Sigma 24 22 1MHz 50usec 18

           

DAC LF 20 20 100kHz 10us 20

DAC RF 14 11 5GHz 500ps 12

12/05/14 CAS2014 ADCs & DACs 31

Application Problems

• Noise, Noise, Noise!– High frequencies

• Use “ground” planes on PCB• Bury HF traces between planes on PCB• Make signal traces differential if possible• Use Common mode chokes• Think HF current paths, minimise length, make differential• Make smooth “flow” Analogue-to-Digital• Remember aliasing and provide adequate filtering

– Additional for Low frequencies• Use “meccas” to control current paths: think current!• Remember 1/f cannot be averaged out• Thermal EMF balance, avoid air flow if possible• Chopper stabilise sensitive input stages

12/05/14 CAS2014 ADCs & DACs 32

Application of fast ADC

• Enhanced Performance with 5 bit DAC– 5 bit 4th order gives excellent performance.

buffer 4 in tegrators Low

Accuracy

5 b it ADC LO G IC

DAC

“5-bit” outputdata

stream

input

5 b itP recise

250 kHz CLK

REF.D ig ita l F ilter

(FPG A)

outputvalue

12/05/14 CAS2014 ADCs & DACs 33

Application Layout

• Think Current!– Uses 20MHz ADC

inside 28 bit loop sensitive to <1uV

– PWM drive has sub ns edges.

– MECCA controls LF currents

– HF currents out & return locally

– PWM differential

– ADC bus return via caps between planes

PWM DIFF.DRIVE

ADC

FPGA

“MECCA”

USB

Input

PW

M S

wit

ch

es

Integrator

ModulationFilter

SIGNAL PLANE

SIGNAL PLANE

POWER SUPPLY PLANE

DIGITAL PLANE

DIGITAL PLANE

CAP

CAP

AD

C

• Using a fast ADC in a Sigma Delta loop

12/05/14 CAS2014 ADCs & DACs 34

Finally…why not use the cryogenics?

• Use Josephson Junction Arrays for the DAC– Future accelerator applications could take advantage of local cryogenics to

use Squids and Josephson arrays.• 15V “SINIS” JJ arrays are being developed now. They could realise the full

potential of the Sigma-Delta ADC architecture.

buffer 4 in tegrators Low

Accuracy

5 b it ADC LO G IC

“5-bit” outputdata

stream

input

JJ A rray5-bit

“slave”

B iasDAC

250 kHz C LK

74 G hzSI SecondReference

D ig ita l F ilter

(FPG A)

outputvalue

12/05/14 CAS2014 ADCs & DACs 35

Conclusions

• Modern “single chip” ADCs, though difficult to use and optimise, offer outstanding performance for the cost.

• NEVER underestimate the difficulties in realising the datasheet promises.• For DC & LF metrology, Delta-Sigma ICs can achieve circa 2 ppm noise and

linearity performance.• A custom designed Delta-Sigma has been incorporated in a Temperature Bridge

and achieves .05ppm linearity.• A SINIS JJ array can be integrated in a Delta-Sigma loop to form a Quantum

Voltmeter with sub ppm resolution.

• Thank you!

• Reference 1: Oversampling Delta-Sigma Data Converters: Theory, Design, and Simulation (Paperback) by James C. Candy (Editor), Gabor C. Temes (Editor) . ISBN 0-87942-285-8 IEEE Press, Wiley. In stock Amazon.